@yoonion/mimi-seed-mcp 0.18.1 → 0.19.0

package/README.md

@@ -99,7 +99,7 @@ export ANTHROPIC_API_KEY=sk-ant-...
 | Android 서명 | 3 | `android_signing_setup` / `android_generate_keystore` / `jenkins_upload_playstore_sa` |
 | 인증 | 4 | `mimi_seed_status` / `mimi_seed_auth_start` / `mimi_seed_auth_status` / `mimi_seed_remote_sync_credentials` |
 | AI (Claude) | 2 | `generate_release_notes_from_commits` / `generate_review_reply` |
-| 영상 제작·YouTube | 14 | `youtube_upload_video` / `youtube_get_video_status` / `youtube_update_video_privacy` / `video_plan_from_story` / `video_render` |
+| 영상 제작·YouTube | 15 | `youtube_upload_video` / `youtube_get_video_status` / `youtube_update_video_privacy` / `video_plan_from_story` / `video_render` |
 
 > 인앱 결제(IAP·구독) 도구는 위 Play Store·App Store 카운트에 포함됩니다 — `appstore_create_inapp_purchase` · `appstore_update_product_review_note` · `appstore_upload_product_review_screenshot` 등.
 > 전체 카탈로그(항상 최신): [`docs/domain/tool-catalog.md`](../../docs/domain/tool-catalog.md)

package/assets/agent-guide.md

@@ -80,7 +80,7 @@ you can paste. Pick the row for the job; batching two rows in one `select:` call
 | CI (GitHub/GitLab) | `select:ci_save_config,ci_list_workflows,ci_trigger_build,ci_get_build_status,ci_list_recent_builds,ci_cancel_build` |
 | Android signing / keystore | `select:android_signing_setup,android_generate_keystore,jenkins_upload_keystore,jenkins_upload_playstore_sa` |
 | Service account end-to-end | `select:iam_list_service_accounts,iam_create_service_account,iam_list_keys,iam_create_key,iam_add_iam_policy_binding,setup_playstore_connection,playstore_register_service_account,playstore_verify_service_account,playstore_list_service_accounts,playstore_delete_service_account` |
-| Story → researched video | `select:video_plan_from_story,video_research_youtube,video_search_stock_assets,video_synthesize_research,video_download_stock_assets,video_generate_image,video_add_local_asset,video_build_timeline,video_render,video_job_status,video_validate` |
+| Story → researched video | `select:video_save_plan,video_plan_from_story,video_research_youtube,video_search_stock_assets,video_synthesize_research,video_download_stock_assets,video_generate_image,video_add_local_asset,video_build_timeline,video_render,video_job_status,video_validate` |
 | YouTube upload / publish | `select:youtube_upload_video,youtube_get_video_status,youtube_update_video_privacy,mimi_seed_auth_start` |
 
 ---
@@ -191,7 +191,7 @@ per-domain inventory is [`docs/domain/tool-catalog.md`](domain/tool-catalog.md).
 | **TikTok Business** | `tiktok_business_auth_status` · `tiktok_business_plan_video_post` · `tiktok_business_publish_video` · `tiktok_business_get_publish_status` |
 | **Checks** | `playstore_check_submission_risks` · `appstore_check_submission_risks` · `screenshot_validate` · `release_status` |
 | **AI / Auth** | `generate_release_notes_from_commits` · `generate_review_reply` · `mimi_seed_status` · `mimi_seed_auth_start` · `mimi_seed_auth_status` · `mimi_seed_remote_sync_credentials` |
-| **Video production** | `youtube_upload_video` · `youtube_get_video_status` · `youtube_update_video_privacy` · `video_plan_from_story` · `video_research_youtube` · `video_search_stock_assets` · `video_synthesize_research` · `video_generate_image` · `video_build_timeline` · `video_render` · `video_job_status` · `video_validate` |
+| **Video production** | `youtube_upload_video` · `youtube_get_video_status` · `youtube_update_video_privacy` · `video_save_plan` · `video_plan_from_story` · `video_research_youtube` · `video_search_stock_assets` · `video_synthesize_research` · `video_generate_image` · `video_build_timeline` · `video_render` · `video_job_status` · `video_validate` |
 
 ---
 
@@ -218,15 +218,16 @@ For visual production, use the bundled `video-create-publish` skill. It requires
 aspect-ratio-specific human-safe crops, video-native motion, an original-resolution frame review, and a saved
 contact sheet; codec validation alone is not a quality pass.
 
-1. `video_plan_from_story` — create the project and scene plan.
+1. `video_save_plan` — save an agent-authored storyboard (no API key; free-path default). `video_plan_from_story` is the metered alternative when `ANTHROPIC_API_KEY` is configured.
 2. `video_research_youtube` — collect reference-only metadata; never treat a result as a render asset.
 3. `video_search_stock_assets` — find licensed Pexels candidates.
 4. `video_synthesize_research` — combine metadata with any direct human/agent observations. Treat its output as
    metadata-bounded guidance, not proof that the source videos were watched.
 5. Preview then confirm `video_download_stock_assets`; use `video_generate_image(confirm=false)` before any paid
-   generation and call it again with `confirm=true` only after approval. User-owned media goes through
-   `video_add_local_asset` with its ownership/license basis.
-6. `video_build_timeline` — provenance-gated scene assignment.
+   generation and call it again with `confirm=true` only after approval. On the free path, generate scene images
+   with the local `codex` CLI (ChatGPT subscription) instead and register them — like all user-owned media —
+   through `video_add_local_asset` with the ownership/license basis.
+6. `video_build_timeline` — provenance-gated scene assignment. Captions burn in the shorts style (bold outline, `**keyword**` highlight); tune with `captionStyle`.
 7. Preview then confirm `video_render`; poll `video_job_status`, then run `video_validate` on the completed MP4.
 
 > **Mimi Seed does not compile app binaries.** It manages metadata, store releases, and

package/dist/registers/video.js

@@ -1,5 +1,5 @@
 import { z } from 'zod';
-import { addLocalAsset, buildTimeline, loadProject, loadTimeline, planStory, } from '../video/project.js';
+import { addLocalAsset, buildTimeline, loadProject, loadTimeline, planStory, savePlan, } from '../video/project.js';
 import { downloadStockAssets, generateImage, researchYouTube, searchStock, } from '../video/providers.js';
 import { getRenderJob, startRender, validateVideo } from '../video/render.js';
 import { synthesizeResearch } from '../video/research.js';
@@ -67,6 +67,29 @@ export function registerVideoTools(server) {
         style: z.string().max(10_000).optional().describe('시각 스타일과 톤'),
         overwrite: z.boolean().default(false).describe('기존 메타데이터를 .history에 보관하고 새 프로젝트로 덮어쓸지 여부'),
     }, async (input) => text(await planStory(input)));
+    server.tool('video_save_plan', [
+        '에이전트가 직접 작성한 장면별 스토리보드를 project.json으로 저장합니다. API 키가 필요 없는 video_plan_from_story의 무료 대안입니다.',
+        '스토리보드(장면 분할·내레이션·화면 문구·시각 프롬프트)는 호출 전에 에이전트가 직접 설계하세요.',
+        '장면 durationSec 합계는 targetDurationSec와 일치해야 합니다(허용 오차 0.1초).',
+        '기존 project.json은 overwrite=true 없이는 덮어쓰지 않습니다.',
+    ].join(' '), {
+        projectDir: absolutePath,
+        title: z.string().min(1).max(500).describe('영상 제목'),
+        story: z.string().min(1).max(200_000).describe('영상으로 만들 story 원문'),
+        language: z.string().default('ko').describe('내레이션과 화면 문구 언어'),
+        aspectRatio: z.enum(['9:16', '16:9', '1:1']).default('9:16').describe('출력 화면비'),
+        targetDurationSec: z.number().min(5).max(300).default(30).describe('목표 영상 길이(초)'),
+        scenes: z.array(z.object({
+            id: z.string().min(1).max(100).optional(),
+            durationSec: z.number().positive().max(300),
+            narration: z.string().max(10_000).optional(),
+            onScreenText: z.string().max(2_000).optional().describe('화면 자막. **단어** 마크업으로 키워드 하이라이트'),
+            visualPrompt: z.string().max(10_000).optional().describe('이미지 생성용 프롬프트'),
+            searchQuery: z.string().max(1_000).optional().describe('스톡 영상 검색용 영문 검색어'),
+        })).min(1).max(12),
+        style: z.string().max(10_000).optional().describe('시각 스타일과 톤'),
+        overwrite: z.boolean().default(false).describe('기존 메타데이터를 .history에 보관하고 새 프로젝트로 덮어쓸지 여부'),
+    }, async (input) => text(savePlan(input)));
     server.tool('video_research_youtube', [
         '스토리보드 검색어나 지정 검색어로 YouTube 유사 영상을 조사합니다.',
         'YOUTUBE_API_KEY가 필요하며 결과는 research/youtube.json에 저장됩니다.',
@@ -157,21 +180,31 @@ export function registerVideoTools(server) {
     server.tool('video_build_timeline', [
         '등록된 이미지·영상 자산을 장면 순서와 길이에 맞춰 timeline.json으로 조합합니다.',
         'assets.json에서 allowedForRendering=true이고 reference-only가 아닌 자산만 허용합니다.',
-        '화면 문구는 최종 렌더에서 자막으로 번인됩니다.',
+        '화면 문구는 최종 렌더에서 쇼츠 스타일 자막(굵은 폰트·흰 글자·검정 외곽선)으로 번인되며, onScreenText 안의 **단어** 마크업은 하이라이트 컬러(기본 노랑)로 강조됩니다.',
+        '한 장면 자막은 3~5어절·최대 2줄로 짧게 쓰고, 하이라이트는 문장당 한 단어만 지정하세요.',
     ].join(' '), {
         projectDir: absolutePath,
         scenes: z.array(z.object({
             id: z.string().min(1),
             assetId: z.string().min(1),
             durationSec: z.number().positive().max(300),
-            onScreenText: z.string().optional(),
+            onScreenText: z.string().optional().describe('번인 자막. **단어**로 키워드 1개 하이라이트, \\n으로 의미 단위 줄바꿈'),
             narration: z.string().optional(),
         })).min(1).max(30),
         audioAssetId: z.string().optional(),
-    }, async ({ projectDir, scenes, audioAssetId }) => text(buildTimeline(projectDir, scenes, audioAssetId)));
+        captionStyle: z.object({
+            preset: z.enum(['shorts', 'box']).optional().describe('shorts(기본): 외곽선 자막+하이라이트 / box: 반투명 박스 자막(튜토리얼용)'),
+            fontName: z.string().min(1).max(200).optional().describe('설치된 굵은 한글 고딕 폰트명(예: Pretendard, 에스코어드림). 생략 시 Malgun Gothic Bold'),
+            fontSizePx: z.number().int().min(24).max(400).optional().describe('생략 시 세로 영상은 높이의 4.2%(1080×1920 기준 81px)'),
+            textColor: z.string().regex(/^#[0-9a-fA-F]{6}$/).optional(),
+            outlineColor: z.string().regex(/^#[0-9a-fA-F]{6}$/).optional(),
+            highlightColor: z.string().regex(/^#[0-9a-fA-F]{6}$/).optional().describe('**단어** 마크업 색. 기본 #FFD400'),
+            position: z.enum(['lower', 'lower-middle', 'center']).optional().describe('생략 시 9:16은 lower-middle(하단-중앙 1/3, 쇼츠 UI 안전영역)'),
+        }).optional().describe('자막 스타일. 생략하면 쇼츠 기본 스타일'),
+    }, async ({ projectDir, scenes, audioAssetId, captionStyle }) => text(buildTimeline(projectDir, scenes, audioAssetId, captionStyle)));
     server.tool('video_render', [
         'timeline.json과 등록 자산을 FFmpeg로 합성해 MP4 렌더 작업을 시작합니다.',
-        '이미지/영상 크롭, 장면 연결, 화면 문구 번인, 선택적 배경음, H.264/yuv420p 출력을 지원합니다.',
+        '이미지/영상 크롭, 장면 연결, 쇼츠 스타일 자막 번인(timeline.json의 captionStyle 적용), 선택적 배경음, H.264/yuv420p 출력을 지원합니다.',
         'confirm=false는 계획만 반환합니다. CPU를 사용하는 로컬 쓰기 작업이므로 승인 후 confirm=true로 호출하세요.',
         '즉시 jobId를 반환하므로 video_job_status로 완료 여부를 확인하세요.',
     ].join(' '), {

package/dist/video/captions.d.ts

@@ -0,0 +1,34 @@
+import type { VideoCaptionStyle, VideoTimeline } from './types.js';
+/**
+ * 쇼츠 자막 스타일 시스템 — SRT + force_style(Arial 18px) 하드코딩을 대체한다.
+ *
+ * 기본 preset 'shorts'는 2026 숏폼 표준(Bold Highlight 스타일)을 따른다:
+ * 굵은 고딕 + 흰 글자 + 검정 외곽선(불투명 박스 없음), 9:16에서는 하단-중앙 1/3 배치,
+ * `**단어**` 마크업으로 문장당 키워드 1개만 하이라이트 컬러.
+ * 'box' preset은 튜토리얼용 반투명 박스 자막(Subtitle Block 스타일)이다.
+ */
+export type CaptionPosition = 'lower' | 'lower-middle' | 'center';
+export interface ResolvedCaptionStyle {
+    fontName: string;
+    fontSizePx: number;
+    textColor: string;
+    outlineColor: string;
+    highlightColor: string;
+    position: CaptionPosition;
+    boxed: boolean;
+}
+export declare function resolveCaptionStyle(width: number, height: number, style?: VideoCaptionStyle): ResolvedCaptionStyle;
+/** #RRGGBB → ASS &HAABBGGRR. alpha 00=불투명, FF=투명. */
+export declare function assColor(hex: string, alpha?: string): string;
+/** ASS 태그·마크업 주입을 막는다. `**`는 하이라이트 마크업으로 살려둔다. */
+declare function sanitizeCaptionText(value: string): string;
+declare function assTime(seconds: number): string;
+/** `**단어**` → 하이라이트 컬러 런. 줄바꿈은 \N. */
+declare function renderEventText(raw: string, style: ResolvedCaptionStyle): string;
+export declare function formatAss(timeline: VideoTimeline, style: ResolvedCaptionStyle): string;
+export declare const __testing: {
+    sanitizeCaptionText: typeof sanitizeCaptionText;
+    assTime: typeof assTime;
+    renderEventText: typeof renderEventText;
+};
+export {};

package/dist/video/captions.js

@@ -0,0 +1,98 @@
+export function resolveCaptionStyle(width, height, style) {
+    const preset = style?.preset ?? 'shorts';
+    const vertical = height > width;
+    return {
+        // Malgun Gothic은 어떤 Windows에도 있는 폴백. 스킬은 설치된 굵은 고딕(Pretendard 등)을 우선 지정한다.
+        fontName: style?.fontName ?? 'Malgun Gothic',
+        // 세로 쇼츠 기준 height의 4.2% ≈ 81px(1080×1920) — 숏폼 권장(55~75pt)과 한국어 예능 자막(72~96px) 사이.
+        fontSizePx: style?.fontSizePx ?? Math.round(height * (vertical ? 0.042 : 0.055)),
+        textColor: style?.textColor ?? '#FFFFFF',
+        outlineColor: style?.outlineColor ?? '#000000',
+        highlightColor: style?.highlightColor ?? '#FFD400',
+        position: style?.position ?? (preset === 'shorts' && vertical ? 'lower-middle' : 'lower'),
+        boxed: preset === 'box',
+    };
+}
+/** #RRGGBB → ASS &HAABBGGRR. alpha 00=불투명, FF=투명. */
+export function assColor(hex, alpha = '00') {
+    const match = /^#([0-9a-fA-F]{2})([0-9a-fA-F]{2})([0-9a-fA-F]{2})$/.exec(hex);
+    if (!match)
+        throw new Error(`잘못된 색상 형식입니다(#RRGGBB): ${hex}`);
+    const [, r, g, b] = match;
+    return `&H${alpha}${b}${g}${r}`.toUpperCase();
+}
+/** ASS 태그·마크업 주입을 막는다. `**`는 하이라이트 마크업으로 살려둔다. */
+function sanitizeCaptionText(value) {
+    return value
+        .replace(/\r/g, '')
+        .replace(/\n{2,}/g, '\n')
+        .replace(/\\/g, '＼')
+        .replace(/</g, '＜')
+        .replace(/>/g, '＞')
+        .replace(/{/g, '｛')
+        .replace(/}/g, '｝')
+        .trim();
+}
+function assTime(seconds) {
+    const cs = Math.max(0, Math.round(seconds * 100));
+    const hours = Math.floor(cs / 360_000);
+    const minutes = Math.floor((cs % 360_000) / 6_000);
+    const secs = Math.floor((cs % 6_000) / 100);
+    const centis = cs % 100;
+    const pad = (value) => String(value).padStart(2, '0');
+    return `${hours}:${pad(minutes)}:${pad(secs)}.${pad(centis)}`;
+}
+/** `**단어**` → 하이라이트 컬러 런. 줄바꿈은 \N. */
+function renderEventText(raw, style) {
+    const primary = assColor(style.textColor);
+    const highlight = assColor(style.highlightColor);
+    return sanitizeCaptionText(raw)
+        .replace(/\*\*([^*\n]+)\*\*/g, `{\\1c${highlight}&}$1{\\1c${primary}&}`)
+        .replace(/\n/g, '\\N');
+}
+export function formatAss(timeline, style) {
+    const { width, height } = timeline;
+    const vertical = height > width;
+    const alignment = style.position === 'center' ? 5 : 2;
+    const marginX = Math.round(width * 0.09);
+    const marginV = style.position === 'center'
+        ? 0
+        : style.position === 'lower-middle'
+            ? Math.round(height * 0.28)
+            : Math.round(height * (vertical ? 0.12 : 0.08));
+    // 한글은 받침 내부 공간이 좁아 외곽선을 폰트 크기의 5.5%로 제한한다. box는 outline 값이 패딩이 된다.
+    const outlinePx = style.boxed
+        ? Math.round(style.fontSizePx * 0.25)
+        : Math.max(2, Math.round(style.fontSizePx * 0.055));
+    const outlineColour = style.boxed ? assColor(style.outlineColor, '50') : assColor(style.outlineColor);
+    const events = [];
+    let cursor = 0;
+    for (const scene of timeline.scenes) {
+        const start = cursor;
+        cursor += scene.durationSec;
+        const text = renderEventText(scene.onScreenText ?? '', style);
+        if (!text)
+            continue;
+        events.push(`Dialogue: 0,${assTime(start)},${assTime(cursor)},Caption,,0,0,0,{\\fad(120,80)}${text}`);
+    }
+    if (events.length === 0)
+        return '';
+    return [
+        '[Script Info]',
+        'ScriptType: v4.00+',
+        `PlayResX: ${width}`,
+        `PlayResY: ${height}`,
+        'WrapStyle: 0',
+        'ScaledBorderAndShadow: yes',
+        '',
+        '[V4+ Styles]',
+        'Format: Name, Fontname, Fontsize, PrimaryColour, SecondaryColour, OutlineColour, BackColour, Bold, Italic, Underline, StrikeOut, ScaleX, ScaleY, Spacing, Angle, BorderStyle, Outline, Shadow, Alignment, MarginL, MarginR, MarginV, Encoding',
+        `Style: Caption,${style.fontName},${style.fontSizePx},${assColor(style.textColor)},${assColor(style.textColor)},${outlineColour},&H80000000,-1,0,0,0,100,100,0,0,${style.boxed ? 3 : 1},${outlinePx},0,${alignment},${marginX},${marginX},${marginV},1`,
+        '',
+        '[Events]',
+        'Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Text',
+        ...events,
+        '',
+    ].join('\n');
+}
+export const __testing = { sanitizeCaptionText, assTime, renderEventText };

package/dist/video/project.d.ts

@@ -1,4 +1,4 @@
-import type { VideoAspectRatio, VideoAsset, VideoAssetKind, VideoAssetManifest, VideoAssetSource, VideoProject, VideoScenePlan, VideoTimeline, VideoTimelineScene } from './types.js';
+import type { VideoAspectRatio, VideoAsset, VideoAssetKind, VideoAssetManifest, VideoAssetSource, VideoCaptionStyle, VideoProject, VideoScenePlan, VideoTimeline, VideoTimelineScene } from './types.js';
 declare function dimensions(aspectRatio: VideoAspectRatio): {
     width: number;
     height: number;
@@ -15,6 +15,23 @@ export interface PlanStoryInput {
     overwrite?: boolean;
 }
 export declare function planStory(input: PlanStoryInput): Promise<VideoProject>;
+export interface SavePlanInput {
+    projectDir: string;
+    title: string;
+    story: string;
+    language: string;
+    aspectRatio: VideoAspectRatio;
+    targetDurationSec: number;
+    scenes: Array<Partial<VideoScenePlan> & Pick<VideoScenePlan, 'durationSec'>>;
+    style?: string;
+    overwrite?: boolean;
+}
+/**
+ * 에이전트(구독 세션)가 직접 작성한 스토리보드를 API 키 없이 project.json으로 저장한다.
+ * video_plan_from_story(ANTHROPIC_API_KEY 필요)의 무료 대안.
+ * 장면 길이 합계가 targetDurationSec와 일치해야 projectSchema 검증을 통과한다.
+ */
+export declare function savePlan(input: SavePlanInput): VideoProject;
 export declare function loadProject(projectDir: string): VideoProject;
 export declare function loadAssets(projectDir: string): VideoAssetManifest;
 export declare function sha256File(filePath: string): string;
@@ -29,7 +46,7 @@ export interface AddLocalAssetInput {
     attribution?: string;
 }
 export declare function addLocalAsset(input: AddLocalAssetInput): VideoAsset;
-export declare function buildTimeline(projectDir: string, scenes: VideoTimelineScene[], audioAssetId?: string): VideoTimeline;
+export declare function buildTimeline(projectDir: string, scenes: VideoTimelineScene[], audioAssetId?: string, captionStyle?: VideoCaptionStyle): VideoTimeline;
 export declare function loadTimeline(projectDir: string): VideoTimeline;
 export declare const __testing: {
     dimensions: typeof dimensions;

package/dist/video/project.js

@@ -90,6 +90,11 @@ export async function planStory(input) {
         },
         scenes: normalizeScenes(parsed.scenes ?? [], input.targetDurationSec),
     }, 'AI가 생성한 영상 프로젝트');
+    persistProject(projectDir, project, input.overwrite === true);
+    return project;
+}
+function persistProject(projectDir, project, overwrite) {
+    const projectPath = path.join(projectDir, PROJECT_FILE);
     mkdirSync(projectDir, { recursive: true });
     mkdirSync(path.join(projectDir, 'assets', 'stock'), { recursive: true });
     mkdirSync(path.join(projectDir, 'assets', 'generated'), { recursive: true });
@@ -97,7 +102,7 @@ export async function planStory(input) {
     mkdirSync(path.join(projectDir, 'research'), { recursive: true });
     mkdirSync(path.join(projectDir, 'render'), { recursive: true });
     mkdirSync(path.join(projectDir, '.jobs'), { recursive: true });
-    if (existsSync(projectPath) && input.overwrite) {
+    if (existsSync(projectPath) && overwrite) {
         const archiveDir = path.join(projectDir, '.history', `${Date.now()}-${project.projectId}`);
         mkdirSync(archiveDir, { recursive: true });
         for (const name of [PROJECT_FILE, ASSET_FILE, TIMELINE_FILE]) {
@@ -115,6 +120,43 @@ export async function planStory(input) {
         projectId: project.projectId,
         assets: [],
     });
+}
+/**
+ * 에이전트(구독 세션)가 직접 작성한 스토리보드를 API 키 없이 project.json으로 저장한다.
+ * video_plan_from_story(ANTHROPIC_API_KEY 필요)의 무료 대안.
+ * 장면 길이 합계가 targetDurationSec와 일치해야 projectSchema 검증을 통과한다.
+ */
+export function savePlan(input) {
+    const projectDir = requireAbsolutePath(input.projectDir, 'projectDir');
+    const projectPath = path.join(projectDir, PROJECT_FILE);
+    if (existsSync(projectPath) && !input.overwrite) {
+        throw new Error(`이미 영상 프로젝트가 있습니다: ${projectPath}\noverwrite=true로 명시해야 덮어씁니다.`);
+    }
+    const dims = dimensions(input.aspectRatio);
+    const project = parseFile(projectSchema, {
+        version: 1,
+        projectId: randomUUID(),
+        title: input.title,
+        story: input.story,
+        language: input.language,
+        createdAt: new Date().toISOString(),
+        settings: {
+            aspectRatio: input.aspectRatio,
+            ...dims,
+            fps: 30,
+            targetDurationSec: input.targetDurationSec,
+            style: input.style,
+        },
+        scenes: input.scenes.map((scene, index) => ({
+            id: scene.id?.trim() || `scene-${index + 1}`,
+            durationSec: scene.durationSec,
+            narration: scene.narration?.trim() ?? '',
+            onScreenText: scene.onScreenText?.trim() ?? '',
+            visualPrompt: scene.visualPrompt?.trim() ?? '',
+            searchQuery: scene.searchQuery?.trim() || scene.visualPrompt?.trim() || '',
+        })),
+    }, '에이전트가 작성한 영상 프로젝트');
+    persistProject(projectDir, project, input.overwrite === true);
     return project;
 }
 export function loadProject(projectDir) {
@@ -191,7 +233,7 @@ export function addLocalAsset(input) {
         allowedForRendering: true,
     });
 }
-export function buildTimeline(projectDir, scenes, audioAssetId) {
+export function buildTimeline(projectDir, scenes, audioAssetId, captionStyle) {
     const dir = requireAbsolutePath(projectDir, 'projectDir');
     const project = loadProject(dir);
     const manifest = loadAssets(dir);
@@ -231,6 +273,7 @@ export function buildTimeline(projectDir, scenes, audioAssetId) {
         totalDurationSec,
         scenes,
         audioAssetId,
+        captionStyle,
     }, '생성할 영상 타임라인');
     writeJsonAtomic(path.join(dir, TIMELINE_FILE), timeline);
     return timeline;

package/dist/video/render.d.ts

@@ -1,9 +1,6 @@
-import type { VideoRenderJob, VideoTimeline } from './types.js';
+import type { VideoRenderJob } from './types.js';
 declare function processAlive(pid: number | undefined): boolean;
 declare function readTail(filePath: string, maxBytes?: number): string;
-declare function sanitizeSubtitleText(value: string): string;
-declare function srtTime(seconds: number): string;
-export declare function formatSrt(timeline: VideoTimeline): string;
 export interface FfmpegPlan {
     args: string[];
     outputPath: string;
@@ -29,9 +26,7 @@ export declare function validateVideo(filePath: string, ffmpegPath?: string): Pr
     issues: string[];
 }>;
 export declare const __testing: {
-    srtTime: typeof srtTime;
     ffprobeFor: typeof ffprobeFor;
-    sanitizeSubtitleText: typeof sanitizeSubtitleText;
     processAlive: typeof processAlive;
     acquireOutputLock: typeof acquireOutputLock;
     readTail: typeof readTail;

package/dist/video/render.js

@@ -3,6 +3,7 @@ import { randomUUID } from 'node:crypto';
 import { closeSync, createWriteStream, existsSync, mkdirSync, openSync, readFileSync, readSync, statSync, unlinkSync, writeFileSync, } from 'node:fs';
 import path from 'node:path';
 import { promisify } from 'node:util';
+import { formatAss, resolveCaptionStyle } from './captions.js';
 import { readJson, writeJsonAtomic } from './files.js';
 import { loadAssets, loadProject, loadTimeline } from './project.js';
 import { parseFile, renderJobSchema } from './schemas.js';
@@ -41,38 +42,6 @@ function readTail(filePath, maxBytes = 64 * 1024) {
     }
     return buffer.toString('utf8');
 }
-function sanitizeSubtitleText(value) {
-    return value
-        .replace(/\r/g, '')
-        .replace(/\n{2,}/g, '\n')
-        .replace(/-->/g, '→')
-        .replace(/</g, '＜')
-        .replace(/>/g, '＞')
-        .replace(/{/g, '｛')
-        .replace(/}/g, '｝')
-        .trim();
-}
-function srtTime(seconds) {
-    const ms = Math.max(0, Math.round(seconds * 1000));
-    const hours = Math.floor(ms / 3_600_000);
-    const minutes = Math.floor((ms % 3_600_000) / 60_000);
-    const secs = Math.floor((ms % 60_000) / 1000);
-    const millis = ms % 1000;
-    return `${String(hours).padStart(2, '0')}:${String(minutes).padStart(2, '0')}:${String(secs).padStart(2, '0')},${String(millis).padStart(3, '0')}`;
-}
-export function formatSrt(timeline) {
-    let cursor = 0;
-    const blocks = [];
-    for (const scene of timeline.scenes) {
-        const text = sanitizeSubtitleText(scene.onScreenText ?? '');
-        const start = cursor;
-        cursor += scene.durationSec;
-        if (!text)
-            continue;
-        blocks.push(`${blocks.length + 1}\n${srtTime(start)} --> ${srtTime(cursor)}\n${text}\n`);
-    }
-    return blocks.join('\n');
-}
 function isImage(asset) {
     return asset.kind === 'image';
 }
@@ -111,14 +80,16 @@ export function buildFfmpegPlan(projectDir, outputPath, jobId) {
         `crop=${project.settings.width}:${project.settings.height},fps=${project.settings.fps},` +
         `trim=duration=${scene.durationSec},setpts=PTS-STARTPTS,format=yuv420p[v${index}]`));
     filters.push(`${timeline.scenes.map((_, index) => `[v${index}]`).join('')}concat=n=${timeline.scenes.length}:v=1:a=0[base]`);
-    const srt = formatSrt(timeline);
+    const captionStyle = resolveCaptionStyle(project.settings.width, project.settings.height, timeline.captionStyle);
+    const ass = formatAss(timeline, captionStyle);
     let videoLabel = '[base]';
     let subtitlePath;
-    if (srt) {
-        const relativeSubtitlePath = `render/captions-${jobId}.srt`;
+    if (ass) {
+        const relativeSubtitlePath = `render/captions-${jobId}.ass`;
         subtitlePath = path.join(dir, relativeSubtitlePath);
-        writeFileSync(subtitlePath, srt, 'utf8');
-        filters.push(`[base]subtitles=${relativeSubtitlePath}:force_style='FontName=Arial,FontSize=18,PrimaryColour=&H00FFFFFF,OutlineColour=&H80000000,BorderStyle=3,Outline=1,Shadow=0,Alignment=2,MarginV=48'[vout]`);
+        writeFileSync(subtitlePath, ass, 'utf8');
+        // 스타일(폰트·크기·외곽선·하이라이트)은 ASS 파일에 내장 — force_style을 쓰지 않는다.
+        filters.push(`[base]subtitles=${relativeSubtitlePath}[vout]`);
         videoLabel = '[vout]';
     }
     args.push('-filter_complex', filters.join(';'), '-map', videoLabel);
@@ -317,4 +288,4 @@ export async function validateVideo(filePath, ffmpegPath) {
         issues.push(`권장 픽셀 포맷 yuv420p가 아닙니다: ${String(video.pix_fmt)}`);
     return { valid: issues.length === 0, format: data.format ?? {}, streams, issues };
 }
-export const __testing = { srtTime, ffprobeFor, sanitizeSubtitleText, processAlive, acquireOutputLock, readTail };
+export const __testing = { ffprobeFor, processAlive, acquireOutputLock, readTail };

package/dist/video/schemas.d.ts

@@ -350,6 +350,32 @@ export declare const assetManifestSchema: z.ZodObject<{
         attribution?: string | undefined;
     }[];
 }>;
+export declare const hexColorSchema: z.ZodString;
+export declare const captionStyleSchema: z.ZodObject<{
+    preset: z.ZodOptional<z.ZodEnum<["shorts", "box"]>>;
+    fontName: z.ZodOptional<z.ZodString>;
+    fontSizePx: z.ZodOptional<z.ZodNumber>;
+    textColor: z.ZodOptional<z.ZodString>;
+    outlineColor: z.ZodOptional<z.ZodString>;
+    highlightColor: z.ZodOptional<z.ZodString>;
+    position: z.ZodOptional<z.ZodEnum<["lower", "lower-middle", "center"]>>;
+}, "strip", z.ZodTypeAny, {
+    preset?: "shorts" | "box" | undefined;
+    fontName?: string | undefined;
+    fontSizePx?: number | undefined;
+    textColor?: string | undefined;
+    outlineColor?: string | undefined;
+    highlightColor?: string | undefined;
+    position?: "lower" | "lower-middle" | "center" | undefined;
+}, {
+    preset?: "shorts" | "box" | undefined;
+    fontName?: string | undefined;
+    fontSizePx?: number | undefined;
+    textColor?: string | undefined;
+    outlineColor?: string | undefined;
+    highlightColor?: string | undefined;
+    position?: "lower" | "lower-middle" | "center" | undefined;
+}>;
 export declare const timelineSceneSchema: z.ZodObject<{
     id: z.ZodString;
     assetId: z.ZodString;
@@ -397,6 +423,31 @@ export declare const timelineSchema: z.ZodEffects<z.ZodObject<{
         onScreenText?: string | undefined;
     }>, "many">;
     audioAssetId: z.ZodOptional<z.ZodString>;
+    captionStyle: z.ZodOptional<z.ZodObject<{
+        preset: z.ZodOptional<z.ZodEnum<["shorts", "box"]>>;
+        fontName: z.ZodOptional<z.ZodString>;
+        fontSizePx: z.ZodOptional<z.ZodNumber>;
+        textColor: z.ZodOptional<z.ZodString>;
+        outlineColor: z.ZodOptional<z.ZodString>;
+        highlightColor: z.ZodOptional<z.ZodString>;
+        position: z.ZodOptional<z.ZodEnum<["lower", "lower-middle", "center"]>>;
+    }, "strip", z.ZodTypeAny, {
+        preset?: "shorts" | "box" | undefined;
+        fontName?: string | undefined;
+        fontSizePx?: number | undefined;
+        textColor?: string | undefined;
+        outlineColor?: string | undefined;
+        highlightColor?: string | undefined;
+        position?: "lower" | "lower-middle" | "center" | undefined;
+    }, {
+        preset?: "shorts" | "box" | undefined;
+        fontName?: string | undefined;
+        fontSizePx?: number | undefined;
+        textColor?: string | undefined;
+        outlineColor?: string | undefined;
+        highlightColor?: string | undefined;
+        position?: "lower" | "lower-middle" | "center" | undefined;
+    }>>;
 }, "strip", z.ZodTypeAny, {
     projectId: string;
     version: 1;
@@ -413,6 +464,15 @@ export declare const timelineSchema: z.ZodEffects<z.ZodObject<{
     }[];
     totalDurationSec: number;
     audioAssetId?: string | undefined;
+    captionStyle?: {
+        preset?: "shorts" | "box" | undefined;
+        fontName?: string | undefined;
+        fontSizePx?: number | undefined;
+        textColor?: string | undefined;
+        outlineColor?: string | undefined;
+        highlightColor?: string | undefined;
+        position?: "lower" | "lower-middle" | "center" | undefined;
+    } | undefined;
 }, {
     projectId: string;
     version: 1;
@@ -429,6 +489,15 @@ export declare const timelineSchema: z.ZodEffects<z.ZodObject<{
     }[];
     totalDurationSec: number;
     audioAssetId?: string | undefined;
+    captionStyle?: {
+        preset?: "shorts" | "box" | undefined;
+        fontName?: string | undefined;
+        fontSizePx?: number | undefined;
+        textColor?: string | undefined;
+        outlineColor?: string | undefined;
+        highlightColor?: string | undefined;
+        position?: "lower" | "lower-middle" | "center" | undefined;
+    } | undefined;
 }>, {
     projectId: string;
     version: 1;
@@ -445,6 +514,15 @@ export declare const timelineSchema: z.ZodEffects<z.ZodObject<{
     }[];
     totalDurationSec: number;
     audioAssetId?: string | undefined;
+    captionStyle?: {
+        preset?: "shorts" | "box" | undefined;
+        fontName?: string | undefined;
+        fontSizePx?: number | undefined;
+        textColor?: string | undefined;
+        outlineColor?: string | undefined;
+        highlightColor?: string | undefined;
+        position?: "lower" | "lower-middle" | "center" | undefined;
+    } | undefined;
 }, {
     projectId: string;
     version: 1;
@@ -461,6 +539,15 @@ export declare const timelineSchema: z.ZodEffects<z.ZodObject<{
     }[];
     totalDurationSec: number;
     audioAssetId?: string | undefined;
+    captionStyle?: {
+        preset?: "shorts" | "box" | undefined;
+        fontName?: string | undefined;
+        fontSizePx?: number | undefined;
+        textColor?: string | undefined;
+        outlineColor?: string | undefined;
+        highlightColor?: string | undefined;
+        position?: "lower" | "lower-middle" | "center" | undefined;
+    } | undefined;
 }>;
 export declare const renderJobSchema: z.ZodObject<{
     id: z.ZodString;

package/dist/video/schemas.js

@@ -56,6 +56,16 @@ export const assetManifestSchema = z.object({
     projectId: z.string().uuid(),
     assets: z.array(assetSchema).max(10_000),
 });
+export const hexColorSchema = z.string().regex(/^#[0-9a-fA-F]{6}$/, '#RRGGBB 형식이어야 합니다.');
+export const captionStyleSchema = z.object({
+    preset: z.enum(['shorts', 'box']).optional(),
+    fontName: z.string().min(1).max(200).optional(),
+    fontSizePx: z.number().int().min(24).max(400).optional(),
+    textColor: hexColorSchema.optional(),
+    outlineColor: hexColorSchema.optional(),
+    highlightColor: hexColorSchema.optional(),
+    position: z.enum(['lower', 'lower-middle', 'center']).optional(),
+});
 export const timelineSceneSchema = z.object({
     id: z.string().min(1).max(100),
     assetId: z.string().min(1).max(200),
@@ -73,6 +83,7 @@ export const timelineSchema = z.object({
     totalDurationSec: z.number().finite().positive().max(300),
     scenes: z.array(timelineSceneSchema).min(1).max(30),
     audioAssetId: z.string().min(1).max(200).optional(),
+    captionStyle: captionStyleSchema.optional(),
 }).superRefine((timeline, ctx) => {
     const calculated = timeline.scenes.reduce((sum, scene) => sum + scene.durationSec, 0);
     if (Math.abs(calculated - timeline.totalDurationSec) > 0.01) {

package/dist/video/types.d.ts

@@ -53,6 +53,15 @@ export interface VideoTimelineScene {
     onScreenText?: string;
     narration?: string;
 }
+export interface VideoCaptionStyle {
+    preset?: 'shorts' | 'box';
+    fontName?: string;
+    fontSizePx?: number;
+    textColor?: string;
+    outlineColor?: string;
+    highlightColor?: string;
+    position?: 'lower' | 'lower-middle' | 'center';
+}
 export interface VideoTimeline {
     version: 1;
     projectId: string;
@@ -63,6 +72,7 @@ export interface VideoTimeline {
     totalDurationSec: number;
     scenes: VideoTimelineScene[];
     audioAssetId?: string;
+    captionStyle?: VideoCaptionStyle;
 }
 export type VideoJobStatus = 'queued' | 'running' | 'completed' | 'failed';
 export interface VideoRenderJob {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@yoonion/mimi-seed-mcp",
-  "version": "0.18.1",
+  "version": "0.19.0",
   "description": "Mimi Seed MCP server \u2014 Firebase + AdMob + Google Play + App Store management for Claude Code / Codex / Cursor / any MCP client.",
   "type": "module",
   "bin": {

package/tool-manifest.json

@@ -1,6 +1,6 @@
 {
   "$comment": "등록된 MCP 도구 + 도메인 메타데이터(label·credential·summary)의 SSOT. src/__tests__/tool-manifest.test.ts 가 실제 서버 등록 목록과 diff 하고 메타데이터 정합성을 검사한다. 도구 추가/삭제/개명 시 이 파일을 함께 갱신할 것 — 산문 문서에는 정확한 개수를 쓰지 말고 이 파일을 가리킬 것. mimi-seed://tools/catalog 리소스가 이 파일을 그대로 서빙한다.",
-  "total": 230,
+  "total": 231,
   "domains": {
     "admob": {
       "label": "AdMob",
@@ -367,6 +367,7 @@
         "youtube_get_video_status",
         "youtube_update_video_privacy",
         "video_plan_from_story",
+        "video_save_plan",
         "video_research_youtube",
         "video_search_stock_assets",
         "video_synthesize_research",